Great question, and your combination of AI and human validation is definitely something that could be strong.
As far as the 'give up' signal, the Agent has several triggers that could stop it from executing, as well as a max number of actions it can take before giving up. Examples of the triggers are page has not loaded after 30s, button clicks or inputs are not responding, etc. If the agent is given the task of 'complete the onboarding process', it will evaluate after each action whether or not it has sufficient information or needs to continue. Once it determines that it has completed the task, it summarizes the actions taken into the final report, including screenshots where applicable.
Are there additional triggers that you think would be helpful? Should the 'max steps' be user-customizable (i.e. maybe you allow it to take up to 50 actions before failing vs. 20 or something like that)
As far as the 'give up' signal, the Agent has several triggers that could stop it from executing, as well as a max number of actions it can take before giving up. Examples of the triggers are page has not loaded after 30s, button clicks or inputs are not responding, etc. If the agent is given the task of 'complete the onboarding process', it will evaluate after each action whether or not it has sufficient information or needs to continue. Once it determines that it has completed the task, it summarizes the actions taken into the final report, including screenshots where applicable.
Are there additional triggers that you think would be helpful? Should the 'max steps' be user-customizable (i.e. maybe you allow it to take up to 50 actions before failing vs. 20 or something like that)
Thank you for your input! It's much appreciated!